biondizzle 3ba41b9322 fix: use slot_token identity check instead of shape heuristic for gather ...

Shape-based check (x_fp4.shape[0] != num_slots) silently fails when
num_tokens == num_slots in L1 (topk=1). Now checks if slot_token is
the identity mapping — only gathers when slot ordering differs from
token ordering.

2026-05-15 10:00:41 +00:00

..

cutlass_nvfp4_gemm

fix: use slot_token identity check instead of shape heuristic for gather

2026-05-15 10:00:41 +00:00

__init__.py

cleanup: remove abandoned TileLang and Mojo files

2026-05-14 12:44:47 +00:00

nvfp4_mega_moe.py

refactor: unify L1/L2 to use 1D slot_expert_ids consistently

2026-05-15 09:56:46 +00:00

symm_buffer.py

Plumb global scale as GEMM alpha instead of folding into UE4M3

2026-05-15 03:32:19 +00:00

weight_transform.py

were getting close

2026-05-15 08:28:40 +00:00